Binary Ninja Headless MCP
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Binary Ninja Headless MCP Disassemble function main in sample.exe"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Binary Ninja Headless MCP
A headless Binary Ninja server that speaks MCP (Model Context Protocol), giving AI agents full access to deep reverse-engineering workflows — disassembly, IL, patching, types, xrefs, and more — without a GUI.
Designed to run in the same Docker container as the agent runtime. No sidecars, no extra services.
This entire project---code, tests, and documentation---is 100% vibe coded.
Why
Existing Binary Ninja MCP servers are either GUI-bound or expose a limited tool surface. This server is headless-only and designed for agent-driven workflows in sandboxed VM/container environments: the agent gets full control over the analysis system, automating large parts of reverse engineering while you interactively discuss and steer the process.
The goal is an interface where agents can inspect, refine, and extend an analysis over time — updating types, symbols, and metadata, improving the analysis database incrementally, applying patches and iterating safely with undo/redo, and running custom scripts when a workflow needs something bespoke.
Related MCP server: IDA Pro MCP Server
Features
181 tools across 36 feature groups: analysis, disassembly, IL, patching, undo/redo, types, workflows, memory, search, xrefs, scripting, and more.
Read-only by default with safe mutation workflows (undo/redo, transactions).
Scripting access via
binja.evalandbinja.callfor anything the tool catalog doesn't cover.Stdio and TCP transports.
Complete command-line client (
binja_cli/binja-cli) with schema-derived flags, JSON batches, offline discovery, and managed persistent servers.Zero runtime dependencies beyond Binary Ninja itself.
Fake backend mode for CI and development without a Binary Ninja license.
Prerequisites
Python
3.11+A Binary Ninja installation with a headless-capable license and the
binaryninjaPython module importable in your runtime (for real analysis)For CI/development without Binary Ninja, use fake backend mode
Installation
git clone https://github.com/mrphrazer/binary-ninja-headless-mcp.git
cd binary-ninja-headless-mcp
pip install .Or install directly from the repo root without cloning:
pip install git+https://github.com/mrphrazer/binary-ninja-headless-mcp.gitQuick Start
Command-line client
The CLI exposes every MCP tool through the same backend. Its usage follows
Ghidra's ghidra_cli. See the complete CLI guide and generated tool reference
for arguments, defaults, sessions, batches, scripting, output, and troubleshooting.
binja_cli list --names-only # Works without Binary Ninja
binja_cli describe session.open
binja_cli server start # Requires licensed Binary Ninja
SID=$(binja_cli --field session_id call session.open --path /bin/ls)
binja_cli --session-id "$SID" call binary.functions --limit 5
binja_cli call session.close --session-id "$SID"
binja_cli server stopUse python3 -m binary_ninja_headless_mcp.binja_cli from a source checkout.
For a license-free client check, run binja_cli --fake-backend call health.ping.
Managed sessions persist across commands; in-process sessions last one invocation.
The managed default is 127.0.0.1:8766. TCP exposes powerful analysis and Python
execution capabilities without authentication; keep it accessible only to trusted clients.
MCP server
Stdio transport (default):
python3 binary_ninja_headless_mcp.pyTCP transport:
python3 binary_ninja_headless_mcp.py --transport tcp --host 127.0.0.1 --port 8765Fake backend mode (no Binary Ninja required):
python3 binary_ninja_headless_mcp.py --fake-backendUse With AI Agents
This server speaks standard MCP over stdio (default) or tcp, so any MCP-capable agent host can use it.
Claude Code
claude mcp add binary_ninja_headless_mcp -- python3 /path/to/binary-ninja-headless-mcp/binary_ninja_headless_mcp.pyOr add it to your project's .mcp.json:
{
"mcpServers": {
"binary_ninja_headless_mcp": {
"command": "python3",
"args": ["binary_ninja_headless_mcp.py"],
"cwd": "/path/to/binary-ninja-headless-mcp"
}
}
}Codex
codex mcp add binary_ninja_headless_mcp -- python3 binary_ninja_headless_mcp.pyGeneric MCP Host
Register a server named
binary_ninja_headless_mcp.Use command
python3with args["binary_ninja_headless_mcp.py"]whencwdis the repo root, or use an absolute script path inargs.Set
cwdto the repo path if you want relative paths likesamples/lsto resolve correctly.Use stdio transport unless your host requires TCP.
For fake mode (no Binary Ninja installed), append
--fake-backend.Verify connectivity by calling
health.ping, thensession.open.
Docker Co-Location Pattern
Recommended deployment model: run the agent process and this MCP server in the same container image.
Example baseline:
FROM python:3.11-slim
WORKDIR /app
COPY . /app
RUN python -m pip install --upgrade pip && pip install ruff pytest
CMD ["python3", "binary_ninja_headless_mcp.py"]If you need real Binary Ninja analysis in-container, add your Binary Ninja runtime + license setup in this same image and start the agent with this MCP server configured.
MCP Methods
initializepingtools/listtools/callshutdown
tools/list behavior:
Without explicit pagination params, returns the full tool catalog.
If
offsetorlimitis provided, uses paginated output (offset=0,limit=50default in paged mode).Supports filtering via:
prefix(for examplebinary.)query(substring match against tool name/description)
Returns pagination metadata:
offset,limit,total,has_more.When a page is truncated (
has_more=true), includesnext_offsetand anoticehint.
Tool call response behavior:
structuredContentis the canonical full payload.content[0].textis a compact summary string (not full JSON duplication).This split is intentional to keep context usage smaller while still exposing full machine-readable data.
Quality And Testing
This repository is well tested and has enforced quality gates.
The CLI verification report records native coverage of all 181 tools across both architectures, transports, and argument styles, along with reproduction commands, installed-package checks, lifecycle evidence, and tested runtime boundaries. The CLI guide documents everyday usage.
Test suite: run
pytest --collect-only -qfor the current collected test count.CI workflow enforces:
ruff format --check .ruff check .pytest
CI uses
BINARY_NINJA_HEADLESS_MCP_FAKE_BACKEND=1so checks run without requiring Binary Ninja installation.Additional structural tests verify tool registry consistency and backend reachability.
Context Controls
Read-only mode is the default for opened sessions (
read_only=true).binary.basic_blocks_atandfunction.basic_blocksare paginated (offset/limit).memory.readhas a hard response cap:length <= 65536.
Limitations
Enterprise APIs are currently not covered.
Debugger APIs are currently not covered.
Security Model
MCP communication (
stdio/tcp) is unauthenticated by default.The server exposes arbitrary scripting via
binja.evaland broad API access viabinja.call.This is by design for trusted, containerized agent environments.
Do not expose this server directly to untrusted users or networks.
Local Dev Workflow
ruff format --check .
ruff check .
BINARY_NINJA_HEADLESS_MCP_FAKE_BACKEND=1 pytest -qFeature Fuzzer
Use the built-in MCP feature fuzzer to exercise a broad tool surface against samples/ls.
Real Binary Ninja backend:
python3 -m binary_ninja_headless_mcp.fuzzer --binary samples/ls --iterations 120 --seed 1337Fake backend smoke run:
python3 -m binary_ninja_headless_mcp.fuzzer --binary samples/ls --fake-backend --allow-tool-errors --iterations 20Write a JSON coverage report:
python3 -m binary_ninja_headless_mcp.fuzzer --binary samples/ls --report-json /tmp/mcp-fuzzer-report.jsonUseful flags:
--min-success-tools N: exits non-zero if fewer thanNtools succeeded.--verbose: print each tool call while fuzzing.--update-analysis: open the seed session withupdate_analysis=true.
Analysis and session lifecycle
session.open, session.open_bytes, and session.open_existing with
update_analysis=true return only after analysis finishes. analysis.update_and_wait
has the same synchronous contract. Neither switches to a task or partial result
on a timer. Sessions are registered before analysis; an analysis failure or
cancellation returns a tool error containing the retained session_id in both
structured data and compact text, so another call can inspect, retry, or close it.
analysis.update now returns a tracked task, just like task.analysis_update.
Poll task.status and fetch task.result when result_ready is true. All terminal
states (completed, failed, cancelled) have a result envelope; failed tasks
include error. A pending result request is a tool error. This is a deliberate
change from the earlier untracked analysis.update response.
Only one managed analysis may be queued, running, or cancelling per session.
Different sessions can analyze concurrently. analysis.status adds status,
start/completion timestamps, last_analysis_task_id, and last_analysis_error to
the existing native state, progress, info, and is_aborted fields. These
managed fields are compatible with Ghidra's lifecycle vocabulary; the native
progress fields remain Binary Ninja specific. has_log is false: there is no
fabricated Ghidra-style analysis log. Native updates triggered by edits or raw API
calls are visible in native progress but are not separate managed tasks.
Cancelling a queued task removes its analysis reservation without aborting the view. Cancelling running analysis requests a native abort and retains ownership until both analysis and abort finish. Once terminal, the session can analyze again; the workflow suspended by our abort is resumed. Cancelling an old terminal task has no effect on a later run. Running search tasks are best effort and drain normally if already started.
An explicit analysis.set_hold or disabled workflow is preserved. Analysis returns
an actionable error while held or disabled; a normal native wait return only counts
as completion when the engine reports idle and is not aborted. Release the hold or
enable the workflow before retrying.
Explicit native control changes invalidate an overlapping managed run even if a
later control restores an idle state. New analysis requests are rejected while
control calls are still in flight. A later explicit workflow disable also
takes precedence over the automatic resume associated with an earlier cancellation.
Binary Ninja can mark interrupted initial analysis as finished before completing
function discovery. On retry, the server re-requests linear sweep when the view's
analysis.linearSweep.autorun setting permits it, preserving its configured
analysis mode. This recovery is retained until a managed analysis completes.
Native heuristics can still produce a different function set after interruption
in reduced analysis modes. Reopen the original input for an independent fresh
analysis when comparing results; completion means the requested native pass
finished, not proof that every possible function was discovered.
loader.rebase requires exclusive use of the session. It rejects active analysis
and other admitted operations; new native calls are rejected during replacement,
while status/list calls use cached metadata. Close waits for replacement to finish.
A rebased view keeps ownership of shared native file metadata, and view-specific
caches are reset. Raw evaluation also protects every session exposed in its context
from concurrent close for the duration of the evaluation.
Use the dedicated lifecycle/control tools for cancellation and rebasing; arbitrary
native calls can bypass managed coordination.
session.close prevents new operations, cancels work, and waits for admitted MCP
calls and background tasks before closing the view. Reopened upload sessions share
ownership of their temporary backing file; it is deleted after the last dependent
session closes, so closing a parent does not break its children.
This ownership is established before a reopened session becomes visible, including
while its initial analysis is still running.
If draining or aborting exceeds 30 seconds, it returns an error, retains the
resources in closing state, and permits a later session.close retry.
analysis.status, session.list, and task metadata remain available. Shutdown
uses one shared 30-second drain deadline; EOF and transport errors also perform
cleanup. The line-delimited stdio transport processes requests sequentially;
use asynchronous tasks for polling on that connection. A second TCP connection
can inspect or cancel a synchronous call on the first connection.
Lifecycle verification
The dedicated stateful driver checks strict results and clean EOF shutdown over an actual stdio subprocess, and fails on unexpected errors or timeouts:
python3 -m binary_ninja_headless_mcp.lifecycle_fuzzer \
--binary samples/ls --seed 1337 --transitions 100 \
--output-dir /tmp/binja-lifecycle-1337
BINJA_NATIVE_TEST_BINARY="$PWD/samples/ls" \
python3 -m pytest tests/test_transport_lifecycle.pyUse a fresh output directory per run. The driver preserves the input, writes a
checkpoint database, incremental request/response trace, progress, stderr, input
hash, engine version, and final report. --timeout sets the external response/task
watchdog (default 300 seconds); a timeout fails verification. To run additional
seeds on a large binary, --analysis-database /path/to/checkpoint.bndb reuses a
completed analysis in a private copy and records the database hash.
--load-options '{"analysis.limits.cacheSize":64}' passes explicit native settings
and records them in the report; neither driver changes product defaults. No native
license is needed for --fake-backend smoke runs or deterministic lifecycle race tests.
The broad feature fuzzer excludes plugin execution and repository modification
from live runs; these actions have tests using doubles. --trace-jsonl preserves
calls incrementally. It reports an unavailable debug parser explicitly. Its default
exit status fails on all unexpected tool errors; --allow-tool-errors is for
exploratory diagnostics and is not a correctness gate. Run it on a disposable copy
when exercising writes. The feature fuzzer also accepts --load-options and
--analysis-database: it copies the database privately and separately probes raw
byte loading without repeating initial analysis.
Feature Catalog
The server currently exposes 181 tools across 36 feature groups.
analysis
analysis.status: Get managed lifecycle status/timestamps/task/error plus native state and progress.analysis.progress: Get analysis progress snapshot.analysis.update: Start tracked analysis; returns task_id. Rejects overlapping runs on this session.analysis.update_and_wait: Wait synchronously until analysis finishes; no timed partial-success response.analysis.abort: Abort analysis.analysis.set_hold: Hold/release analysis queue.
annotation
annotation.rename_function: Rename a function.annotation.rename_symbol: Rename symbol at address.annotation.undefine_symbol: Undefine user symbol at address.annotation.define_symbol: Define symbol at address.annotation.rename_data_var: Rename data variable.annotation.define_data_var: Define data variable.annotation.undefine_data_var: Undefine data variable.annotation.set_comment: Set comment at address.annotation.get_comment: Get comment at address.annotation.add_tag: Add user data tag at address.annotation.get_tags: Get tags at address.
arch
arch.info: Get architecture and platform metadata.arch.disasm_bytes: Disassemble bytes with selected architecture.arch.assemble: Assemble instruction text with selected architecture.
baseaddr
baseaddr.detect: Run base-address detection.baseaddr.reasons: Get base-address detection reasons.baseaddr.abort: Abort base-address detection.
binary
binary.summary: Get binary/session summary.binary.save: Save the current binary view to a file path.binary.functions: List functions with pagination.binary.strings: List discovered strings with pagination.binary.search_text: Search raw text/bytes in a session.binary.sections: List sections with pagination.binary.segments: List segments with pagination.binary.symbols: List symbols with pagination.binary.data_vars: List data variables with pagination.binary.get_function_at: Find function by address.binary.get_function_disassembly_at: Get full disassembly for function containing an address.binary.get_function_il_at: Get full IL for function containing an address.binary.functions_at: List functions at an address.binary.basic_blocks_at: List basic blocks at an address with pagination.
binja
binja.info: Return Binary Ninja version/install info.binja.call: Generic API bridge: callbn.*orbv.*target path.binja.eval: Evaluate Python code withbn,sessions, and optionalbv.
data
data.typed_at: Get typed data variable at an address.
database
database.create_bndb: Create .bndb from session.database.save_auto_snapshot: Save auto snapshot.database.info: Get database status for session.database.snapshots: List database snapshots.database.read_global: Read database global string key.database.write_global: Write database global string key.
debug
debug.parsers: List debug info parsers valid for this view.debug.parse_and_apply: Parse debug info and apply it to the view.
disasm
disasm.linear: Get linear disassembly lines.disasm.function: Get full disassembly for function containing an address.disasm.range: Address-range disassembly lines.
external
external.library_add: Add external library.external.library_list: List external libraries.external.library_remove: Remove external library.external.location_add: Add external location mapping.external.location_get: Get external location mapping.external.location_remove: Remove external location mapping.
function
function.basic_blocks: List basic blocks in a function with pagination.function.callers: Callers of a function.function.callees: Callees of a function.function.variables: List function variables.function.var_refs: List variable references in MLIL/HLIL.function.var_refs_from: List variable references originating from an address.function.ssa_var_def_use: Get SSA variable definition and uses.function.ssa_memory_def_use: Get SSA memory definition and uses by memory version.function.metadata_store: Store function metadata by key.function.metadata_query: Query function metadata by key.function.metadata_remove: Remove function metadata by key.
health
health.ping: Health check.
il
il.function: IL function listing.il.instruction_by_addr: Get IL instruction by source address.il.address_to_index: Map address to IL index/indices.il.index_to_address: Map IL index to source address.il.rewrite.capabilities: List IL rewrite support for one function and IL level.il.rewrite.noop_replace: Perform no-op IL expression replacement.il.rewrite.translate_identity: Translate IL with identity mapping callback.
loader
loader.rebase: Rebase BinaryView.loader.load_settings_types: List loader settings type names.loader.load_settings_get: Get loader settings values.loader.load_settings_set: Set one loader setting value.
memory
memory.read: Read bytes from the view (length <= 65536).memory.write: Write bytes (hex) to the view.memory.insert: Insert bytes (hex) into the view.memory.remove: Remove bytes from the view.memory.reader_read: Read integer values via BinaryReader.memory.writer_write: Write integer values via BinaryWriter.
mcp
mcp.response_format: Explain tool result fields (structuredContentfull payload,content[0].textsummary).
metadata
metadata.store: Store metadata by key.metadata.query: Query metadata by key.metadata.remove: Remove metadata by key.
patch
patch.assemble: Assemble and patch instruction bytes at address.patch.status: Inspect patch availability at address.patch.convert_to_nop: Patch instruction to NOP when supported.patch.always_branch: Patch conditional branch to always branch when supported.patch.never_branch: Patch conditional branch to never branch when supported.patch.invert_branch: Patch conditional branch by inversion when supported.patch.skip_and_return_value: Patch instruction to skip and return value when supported.
plugin
plugin.valid_commands: List context-valid plugin commands.plugin.execute: Execute a context-valid plugin command.
plugin_repo
plugin_repo.status: List plugin repositories and plugin states.plugin_repo.check_updates: Check plugin repository updates.plugin_repo.plugin_action: Run install/uninstall/enable/disable action on repository plugin.
project
project.create: Create project.project.open: Open project.project.close: Close tracked project.project.list: List project folders/files.project.create_folder: Create project folder.project.create_file: Create project file from base64 data.project.metadata_store: Store project metadata.project.metadata_query: Query project metadata.project.metadata_remove: Remove project metadata.
search
search.data: Search for raw byte patterns (hex string).search.next_text: Find next text match.search.all_text: Find all text matches in range (regex optional).search.next_data: Find next data/byte-pattern match.search.all_data: Find all data/byte-pattern matches in range.search.next_constant: Find next constant occurrence.search.all_constant: Find all constant occurrences in range.
section
section.add_user: Add user section.section.remove_user: Remove user section.
segment
segment.add_user: Add user segment.segment.remove_user: Remove user segment.
session
session.open: Open a binary and create a session.session.open_bytes: Open a binary session from base64-encoded bytes.session.open_existing: Open another session from an existing session's file.session.close: Close one open session.session.list: List open sessions.session.mode: Get session safety/determinism mode.session.set_mode: Update session safety/determinism mode.
task
task.analysis_update: Alias of analysis.update: start tracked analysis and return task_id.task.search_text: Start async search task.task.status: Get task status.task.result: Get a completed, failed, or cancelled result envelope; pending tasks return an error.task.cancel: Request cancellation. Terminal tasks are unchanged; running analysis drains before reuse.
transform
transform.inspect: Inspect/process transform extraction pipeline. For a session input, the root context is retained in the selected contexts so temporary transform cleanup preserves the session's view.
type
type.parse_string: Parse a single type string.type.parse_declarations: Parse C declarations for types/variables/functions.type.define_user: Define user type from type source.type.rename: Rename a type.type.undefine_user: Undefine a user type.type.import_library_type: Import type from type library.type.import_library_object: Import object type from type library.type.export_to_library: Export type into a type library.
type_archive
type_archive.create: Create and optionally attach a type archive.type_archive.open: Open and optionally attach a type archive.type_archive.list: List attached type archives.type_archive.get: Get one tracked type archive.type_archive.pull: Pull types from a type archive.type_archive.push: Push types to a type archive.type_archive.references: Query archive incoming/outgoing references for one type.
type_library
type_library.create: Create and optionally attach a type library.type_library.load: Load and optionally attach a type library.type_library.list: List type libraries attached to the view.type_library.get: Get one tracked type library.
uidf
uidf.parse_possible_value: Parse user-informed possible value set string.uidf.set_user_var_value: Set function user variable value.uidf.clear_user_var_value: Clear function user variable value.uidf.list_user_var_values: List all user variable values for a function.
undo
undo.begin: Begin undo transaction.undo.commit: Commit undo transaction.undo.revert: Revert undo transaction.undo.undo: Perform undo.undo.redo: Perform redo.
value
value.reg: Get register value at/after an address.value.stack: Get stack contents at/after an address.value.possible: Get IL possible value set at an address.value.flags_at: Get lifted IL flag read/write state at an address.
workflow
workflow.list: List registered workflows.workflow.describe: Describe workflow topology and settings.workflow.clone: Clone workflow.workflow.insert: Insert activities before an activity.workflow.insert_after: Insert activities after an activity.workflow.remove: Remove workflow activity.workflow.graph: Summarize workflow graph.workflow.machine.status: Get workflow machine status.workflow.machine.control: Control workflow machine runtime.
xref
xref.code_refs_to: Code references to an address.xref.code_refs_from: Code references from an address.xref.data_refs_to: Data references to an address.xref.data_refs_from: Data references from an address.
Contact
For more information, contact Tim Blazytko (@mr_phrazer).
Available Tools
181 toolsanalysis.abortC
Abort analysis.
Inputs: session_id (required).
analysis.progressC
Get analysis progress snapshot.
Inputs: session_id (required).
analysis.set_holdC
Hold/release analysis queue.
Inputs: hold (required), session_id (required).
analysis.statusC
Get analysis status.
Inputs: session_id (required).
analysis.updateC
Trigger async analysis update.
Inputs: session_id (required).
analysis.update_and_waitC
Run analysis update and wait for completion.
Inputs: session_id (required).
annotation.add_tagC
Add user data tag at address.
Inputs: data (required), address (required), tag_type (required), session_id (required), +1 more.
annotation.define_data_varD
Define data variable.
Inputs: address (required), session_id (required), name, width, +1 more.
annotation.define_symbolC
Define symbol at address.
Inputs: name (required), address (required), session_id (required), symbol_type.
annotation.get_commentC
Get comment at address.
Inputs: address (required), session_id (required).
annotation.get_tagsC
Get tags at address.
Inputs: address (required), session_id (required).
annotation.rename_data_varD
Rename data variable.
Inputs: address (required), new_name (required), session_id (required).
annotation.rename_functionD
Rename a function.
Inputs: new_name (required), session_id (required), function_start (required).
annotation.rename_symbolC
Rename symbol at address.
Inputs: address (required), new_name (required), session_id (required).
annotation.set_commentC
Set comment at address.
Inputs: address (required), comment (required), session_id (required).
annotation.undefine_data_varD
Undefine data variable.
Inputs: address (required), session_id (required).
annotation.undefine_symbolC
Undefine user symbol at address.
Inputs: address (required), session_id (required).
arch.assembleC
Assemble instruction text with selected architecture.
Inputs: asm (required), session_id (required), address, arch_name.
arch.disasm_bytesC
Disassemble bytes with selected architecture.
Inputs: data_hex (required), session_id (required), address, arch_name.
arch.infoC
Get architecture and platform metadata.
Inputs: session_id (required).
baseaddr.abortC
Abort base-address detection.
Inputs: session_id (required).
baseaddr.detectD
Run base-address detection.
Inputs: session_id (required), analysis, alignment, arch_name, +4 more.
baseaddr.reasonsC
Get base-address detection reasons.
Inputs: session_id (required), base_address (required).
binary.basic_blocks_atB
List basic blocks at an address with pagination.
Inputs: address (required), session_id (required), limit, offset.
binary.data_varsC
List data variables with pagination.
Inputs: session_id (required), limit, offset.
binary.functionsC
List functions with pagination.
Inputs: session_id (required), limit, offset.
binary.functions_atC
List functions at an address.
Inputs: address (required), session_id (required).
binary.get_function_atB
Find function by address.
Inputs: address (required), session_id (required).
binary.get_function_disassembly_atC
Get full disassembly for the function containing an address.
Inputs: address (required), session_id (required).
binary.get_function_il_atC
Get full IL for the function containing an address.
Inputs: address (required), session_id (required), ssa, level.
binary.saveC
Save the current binary view to a file path.
Inputs: path (required), session_id (required).
binary.search_textC
Search raw text/bytes in a session.
Inputs: query (required), session_id (required), limit.
binary.sectionsC
List sections with pagination.
Inputs: session_id (required), limit, offset.
binary.segmentsC
List segments with pagination.
Inputs: session_id (required), limit, offset.
binary.stringsC
List discovered strings with pagination.
Inputs: session_id (required), limit, offset.
binary.summaryC
Get binary/session summary.
Inputs: session_id (required).
binary.symbolsC
List symbols with pagination.
Inputs: session_id (required), limit, offset.
binja.callC
Generic API bridge: call bn.* or bv.* target path.
Inputs: target (required), args, kwargs, session_id.
binja.evalC
Evaluate Python code with bn, sessions, and optional bv.
Inputs: code (required), session_id.
binja.infoA
Return Binary Ninja version/install info.
Inputs: no named parameters.
database.create_bndbC
Create .bndb from session.
Inputs: path (required), session_id (required).
database.infoC
Get database status for session.
Inputs: session_id (required).
database.read_globalC
Read database global string key.
Inputs: key (required), session_id (required).
database.save_auto_snapshotD
Save auto snapshot.
Inputs: session_id (required).
database.snapshotsC
List database snapshots.
Inputs: session_id (required), limit, offset.
database.write_globalC
Write database global string key.
Inputs: key (required), value (required), session_id (required).
data.typed_atD
Get typed data variable at an address.
Inputs: address (required), session_id (required).
debug.parse_and_applyC
Parse debug info and apply it to the view.
Inputs: session_id (required), debug_path, parser_name.
debug.parsersC
List debug info parsers valid for this view.
Inputs: session_id (required).
disasm.functionC
Get full disassembly for the function containing an address.
Inputs: address (required), session_id (required).
disasm.linearD
Get linear disassembly lines.
Inputs: session_id (required), limit, offset.
disasm.rangeC
Address-range disassembly lines.
Inputs: start (required), length (required), session_id (required), limit.
external.library_addC
Add external library.
Inputs: name (required), session_id (required), auto.
external.library_listD
List external libraries.
Inputs: session_id (required).
external.library_removeC
Remove external library.
Inputs: name (required), session_id (required).
external.location_addD
Add external location mapping.
Inputs: session_id (required), source_address (required), auto, library_name, +2 more.
external.location_getC
Get external location mapping.
Inputs: session_id (required), source_address (required).
external.location_removeC
Remove external location mapping.
Inputs: session_id (required), source_address (required).
function.basic_blocksC
List basic blocks in a function with pagination.
Inputs: session_id (required), function_start (required), limit, offset.
function.calleesD
Callees of a function.
Inputs: session_id (required), function_start (required).
function.callersD
Callers of a function.
Inputs: session_id (required), function_start (required).
function.metadata_queryC
Query function metadata by key.
Inputs: key (required), session_id (required), function_start (required).
function.metadata_removeC
Remove function metadata by key.
Inputs: key (required), session_id (required), function_start (required).
function.metadata_storeC
Store function metadata by key.
Inputs: key (required), value (required), session_id (required), function_start (required).
function.ssa_memory_def_useC
Get SSA memory definition and uses by memory version.
Inputs: version (required), session_id (required), function_start (required), level.
function.ssa_var_def_useC
Get SSA variable definition and uses.
Inputs: version (required), session_id (required), variable_name (required), function_start (required), +1 more.
function.variablesC
List function variables.
Inputs: session_id (required), function_start (required).
function.var_refsC
List variable references in MLIL/HLIL.
Inputs: session_id (required), variable_name (required), function_start (required), level.
function.var_refs_fromC
List variable references originating from an address.
Inputs: address (required), session_id (required), function_start (required), level, +1 more.
health.pingA
Health check.
Inputs: no named parameters.
il.address_to_indexC
Map address to IL index/indices.
Inputs: address (required), session_id (required), function_start (required), ssa, +1 more.
il.functionD
IL function listing.
Inputs: session_id (required), function_start (required), ssa, level, +2 more.
il.index_to_addressC
Map IL index to source address.
Inputs: index (required), session_id (required), function_start (required), ssa, +1 more.
il.instruction_by_addrC
Get IL instruction by source address.
Inputs: address (required), session_id (required), function_start (required), ssa, +1 more.
il.rewrite.capabilitiesC
List IL rewrite support for one function and IL level.
Inputs: session_id (required), function_start (required), level.
il.rewrite.noop_replaceC
Perform no-op IL expression replacement.
Inputs: session_id (required), function_start (required), index, level, +2 more.
il.rewrite.translate_identityD
Translate IL with identity mapping callback.
Inputs: session_id (required), function_start (required), level.
loader.load_settings_getC
Get loader settings values.
Inputs: type_name (required), session_id (required).
loader.load_settings_setC
Set one loader setting value.
Inputs: key (required), value (required), type_name (required), session_id (required), +1 more.
loader.load_settings_typesC
List loader settings type names.
Inputs: session_id (required).
loader.rebaseD
Rebase BinaryView.
Inputs: address (required), session_id (required), force.
mcp.response_formatA
Explain MCP tool result fields (structuredContent full payload, content[0].text summary).
Inputs: no named parameters.
memory.insertC
Insert bytes (hex) into the view.
Inputs: address (required), data_hex (required), session_id (required).
memory.readC
Read bytes from the view.
Inputs: length (required), address (required), session_id (required).
memory.reader_readD
Read integer values via BinaryReader.
Inputs: width (required), address (required), session_id (required), endian.
memory.removeC
Remove bytes from the view.
Inputs: length (required), address (required), session_id (required).
memory.writeC
Write bytes (hex) to the view.
Inputs: address (required), data_hex (required), session_id (required).
memory.writer_writeD
Write integer values via BinaryWriter.
Inputs: value (required), width (required), address (required), session_id (required), +1 more.
metadata.queryC
Query metadata by key.
Inputs: key (required), session_id (required).
metadata.removeC
Remove metadata by key.
Inputs: key (required), session_id (required).
metadata.storeD
Store metadata by key.
Inputs: key (required), value (required), session_id (required).
patch.always_branchC
Patch conditional branch to always branch when supported.
Inputs: address (required), session_id (required).
patch.assembleC
Assemble and patch instruction bytes at address.
Inputs: asm (required), address (required), session_id (required).
patch.convert_to_nopC
Patch instruction to NOP when supported.
Inputs: address (required), session_id (required).
patch.invert_branchC
Patch conditional branch by inversion when supported.
Inputs: address (required), session_id (required).
patch.never_branchC
Patch conditional branch to never branch when supported.
Inputs: address (required), session_id (required).
patch.skip_and_return_valueD
Patch instruction to skip and return value when supported.
Inputs: value (required), address (required), session_id (required).
patch.statusC
Inspect patch availability at address.
Inputs: address (required), session_id (required).
plugin.executeD
Execute a context-valid plugin command.
Inputs: name (required), session_id (required), length, address, +1 more.
plugin_repo.check_updatesD
Check plugin repository updates.
Inputs: perform.
plugin_repo.plugin_actionC
Run install/uninstall/enable/disable action on repository plugin.
Inputs: action (required), plugin_path (required), repository_path (required).
plugin_repo.statusB
List plugin repositories and plugin states.
Inputs: no named parameters.
plugin.valid_commandsC
List context-valid plugin commands.
Inputs: session_id (required), length, address.
project.closeC
Close tracked project.
Inputs: project_id (required).
project.createD
Create project.
Inputs: name (required), path (required).
project.create_fileC
Create project file from base64 data.
Inputs: name (required), project_id (required), data_base64 (required), folder_id, +1 more.
project.create_folderD
Create project folder.
Inputs: name (required), project_id (required), description, parent_folder_id.
project.listC
List project folders/files.
Inputs: project_id (required).
project.metadata_queryD
Query project metadata.
Inputs: key (required), project_id (required).
project.metadata_removeC
Remove project metadata.
Inputs: key (required), project_id (required).
project.metadata_storeC
Store project metadata.
Inputs: key (required), value (required), project_id (required).
project.openC
Open project.
Inputs: path (required).
search.all_constantC
Find all constant occurrences in range.
Inputs: end (required), start (required), constant (required), session_id (required), +1 more.
search.all_dataC
Find all data/byte-pattern matches in range.
Inputs: end (required), start (required), data_hex (required), session_id (required), +1 more.
search.all_textC
Find all text matches in range (regex optional).
Inputs: end (required), query (required), start (required), session_id (required), +2 more.
search.dataC
Search for raw byte patterns (hex string).
Inputs: data_hex (required), session_id (required), end, limit, +1 more.
search.next_constantC
Find next constant occurrence.
Inputs: start (required), constant (required), session_id (required).
search.next_dataC
Find next data/byte-pattern match.
Inputs: start (required), data_hex (required), session_id (required).
search.next_textD
Find next text match.
Inputs: query (required), start (required), session_id (required).
section.add_userD
Add user section.
Inputs: name (required), start (required), length (required), session_id (required), +4 more.
section.remove_userD
Remove user section.
Inputs: name (required), session_id (required).
segment.add_userD
Add user segment.
Inputs: start (required), length (required), session_id (required), readable, +6 more.
segment.remove_userC
Remove user segment.
Inputs: start (required), session_id (required), length.
session.closeC
Close one open session.
Inputs: session_id (required).
session.listB
List open sessions.
Inputs: no named parameters.
session.modeA
Get session safety/determinism mode.
Inputs: session_id (required).
session.openC
Open a binary and create a session.
Inputs: path (required), options, read_only, deterministic, +1 more.
session.open_bytesC
Open a binary session from base64-encoded bytes.
Inputs: data_base64 (required), options, filename, read_only, +2 more.
session.open_existingC
Open another session from an existing session's file.
Inputs: source_session_id (required), options, read_only, deterministic, +1 more.
session.set_modeC
Update session safety/determinism mode.
Inputs: session_id (required), read_only, deterministic.
task.analysis_updateC
Start async analysis update task.
Inputs: session_id (required).
task.cancelC
Cancel task (best-effort).
Inputs: task_id (required).
task.resultC
Get task result.
Inputs: task_id (required).
task.search_textC
Start async search task.
Inputs: query (required), session_id (required), limit.
task.statusC
Get task status.
Inputs: task_id (required).
transform.inspectD
Inspect/process transform pipeline. Requires session_id or path.
Inputs: mode, path, process, session_id.
type_archive.createC
Create and optionally attach a type archive.
Inputs: path (required), session_id (required), attach, platform_name.
type_archive.getC
Get one tracked type archive.
Inputs: session_id (required), type_archive_id (required).
type_archive.listC
List attached type archives.
Inputs: session_id (required).
type_archive.openC
Open and optionally attach a type archive.
Inputs: path (required), session_id (required), attach.
type_archive.pullD
Pull types from a type archive.
Inputs: names (required), session_id (required), type_archive_id (required).
type_archive.pushC
Push types to a type archive.
Inputs: names (required), session_id (required), type_archive_id (required).
type_archive.referencesC
Query archive incoming/outgoing references for one type.
Inputs: name (required), type_archive_id (required).
type.define_userD
Define user type from type source.
Inputs: session_id (required), type_source (required), name, import_dependencies.
type.export_to_libraryC
Export type into a type library.
Inputs: session_id (required), type_source (required), type_library_id (required), name, +1 more.
type.import_library_objectC
Import object type from type library.
Inputs: name (required), session_id (required), type_library_id.
type.import_library_typeC
Import type from type library.
Inputs: name (required), session_id (required), type_library_id.
type_library.createC
Create and optionally attach a type library.
Inputs: name (required), session_id (required), path, add_to_view.
type_library.getC
Get one tracked type library.
Inputs: session_id (required), type_library_id (required).
type_library.listC
List type libraries attached to the view.
Inputs: session_id (required).
type_library.loadC
Load and optionally attach a type library.
Inputs: path (required), session_id (required), add_to_view.
type.parse_declarationsC
Parse C declarations for types/variables/functions.
Inputs: session_id (required), declarations (required), options, include_dirs, +1 more.
type.parse_stringD
Parse a single type string.
Inputs: session_id (required), type_source (required), import_dependencies.
type.renameD
Rename a type.
Inputs: new_name (required), old_name (required), session_id (required).
type.undefine_userD
Undefine a user type.
Inputs: name (required), session_id (required).
uidf.clear_user_var_valueC
Clear function user variable value.
Inputs: def_addr (required), session_id (required), variable_name (required), function_start (required), +1 more.
uidf.list_user_var_valuesC
List all user variable values for a function.
Inputs: session_id (required), function_start (required).
uidf.parse_possible_valueC
Parse user-informed possible value set string.
Inputs: state (required), value (required), session_id (required), here.
uidf.set_user_var_valueC
Set function user variable value.
Inputs: state (required), value (required), def_addr (required), session_id (required), +3 more.
undo.beginC
Begin undo transaction.
Inputs: session_id (required).
undo.commitC
Commit undo transaction.
Inputs: session_id (required), transaction_id.
undo.redoD
Perform redo.
Inputs: session_id (required).
undo.revertD
Revert undo transaction.
Inputs: session_id (required), transaction_id.
undo.undoD
Perform undo.
Inputs: session_id (required).
value.flags_atC
Get lifted IL flag read/write state at an address.
Inputs: address (required), session_id (required), function_start (required).
value.possibleC
Get IL possible value set at an address.
Inputs: address (required), session_id (required), function_start (required), ssa, +1 more.
value.regC
Get register value at/after an address.
Inputs: address (required), register (required), session_id (required), function_start (required), +1 more.
value.stackC
Get stack contents at/after an address.
Inputs: size (required), address (required), session_id (required), stack_offset (required), +2 more.
workflow.cloneD
Clone workflow.
Inputs: name (required), session_id (required), register.
workflow.describeC
Describe workflow topology and settings.
Inputs: session_id (required), activity, immediate, workflow_name.
workflow.graphD
Summarize workflow graph.
Inputs: session_id (required), activity, sequential, workflow_name.
workflow.insertC
Insert activities before an activity.
Inputs: activity (required), activities (required), session_id (required), workflow_name.
workflow.insert_afterC
Insert activities after an activity.
Inputs: activity (required), activities (required), session_id (required), workflow_name.
workflow.listA
List registered workflows.
Inputs: no named parameters.
workflow.machine.controlD
Control workflow machine runtime.
Inputs: action (required), session_id (required), enable, activity, +4 more.
workflow.machine.statusC
Get workflow machine status.
Inputs: session_id (required), workflow_name.
workflow.removeC
Remove workflow activity.
Inputs: activity (required), session_id (required), workflow_name.
xref.code_refs_fromC
Code references from an address.
Inputs: address (required), session_id (required), length.
xref.code_refs_toC
Code references to an address.
Inputs: address (required), session_id (required), limit, offset.
xref.data_refs_fromD
Data references from an address.
Inputs: address (required), session_id (required), length.
xref.data_refs_toB
Data references to an address.
Inputs: address (required), session_id (required), limit.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 181 tools
While the use of prefixes (e.g., session., binary., function.) helps organize tools, several pairs have nearly identical descriptions (e.g., binary.get_function_disassembly_at and disasm.function both say 'Get full disassembly for the function containing an address'), creating confusion. The high number of tools (181) also makes it harder to distinguish subtle differences.
All tools follow a consistent dot-separated naming convention with lower-case words, typically a prefix indicating the domain followed by a descriptive verb or noun (e.g., session.open, binary.functions, annotation.rename_function). No mixing of styles or irregularities.
At 181 tools, this server drastically exceeds the typical MCP scope (3-15 tools). Even for a comprehensive binary analysis platform, this count is extreme and makes the tool set unwieldy for agents. The server would benefit from being split into multiple focused servers.
The tool set covers an exhaustive range of operations for binary analysis: session management, analysis control, memory, disassembly, IL, annotations, patching, undo, tasks, types, workflows, projects, plugins, and more. There are no obvious gaps for the headless Binary Ninja use case.
Related MCP Connectors
Develop, manage, and debug Railway projects, services, and deployments from within agents.
CVE lookups (NVD) and dependency-manifest audits (OSV) for AI agents. No API keys.
CVE lookups (NVD) and dependency-manifest audits (OSV) for AI agents. No API keys.
Build, run, inspect, and continue AgentScript workflows on connected Super Chrome and Mac clients.
Related MCP Servers
- AlicenseNot gradedqualityBmaintenanceEnables AI agents to inspect, reverse engineer, decompile, and operate on binaries through IDA Pro, including headless and GUI sessions, over stdio or Streamable HTTP.MIT
- FlicenseNot gradedqualityAmaintenanceEnables AI assistants to remotely control a disassembler/debugger for reverse engineering, including executing Python scripts, inspecting binaries, searching for values, text, or byte sequences, and retrieving disassembly, functions, exports, and strings.-
- FlicenseNot gradedqualityAmaintenanceEnables AI assistants to automate reverse engineering and binary analysis by running scripts, retrieving metadata, searching values, text, and byte sequences, and reading disassembly.-
- FlicenseNot gradedqualityAmaintenanceEnables AI agents to import, disassemble, and debug binaries directly so they can perform cross-reference reasoning, rename stripped functions, and recover structure layouts during native reverse-engineering workflows. It integrates with offline ELF reconnaissance output to support symbol and struct recovery without opening the Ghidra GUI.-